Defiler

Terrible SCSI performance in Windows XP

Recommended Posts

I found out why the benchmarks on my old XP installation were so bad (SCSI and IDE, see about 20 posts up there...). When I shut down the distributed.net client on that system, the scores go up to the level of the clean XP installation. However, it is no cure for the SCSI Problem.

The dnet client does the same on my W2k installation. So, if anyone has low hd performance with SCSI and IDE drives and has the dnet client running, this might be worth a shot.

Share this post


Link to post
Share on other sites

As with it seems a few people I have been following this thread. Also because it was mention the MS research peeps are possibly looking here I thought I would submit my info as well. I also apologise for the length, but I hope it proves useful :wink:

System Specs:

AMD XP 2000+

ABit KR7A-Raid

ABit Siluro GeForce4 Ti4600

2x Kingmax PC2700

Adaptec 39160

Atlas 10k2 36.5gb

2x Seagate 18.5gb Ultra Wide 2 10krpm in software Raid 0

Ok here is the Atlas10k2 benchmark, showing first Win2k and then WinXP:

atto-benches.jpg

(the win2k reults were saved to a Atto file)

Obviously this is showing the same trend as other people, lets all pray to the hotfix gods to get this sorted ASAP. I like WinXP to much to have to go back to Win2k.

Now here my interesting results. My harddrive config was intended as follows. Boot from the U160 Atlas10k2 and use the 2 older Seagate U2W drives for the rest of my storage. In effect almost having two identically performing drives (without the WinXP effect). I only sorted out the software Raid 0 the other day and it was my first attempt at Raid anything.

Windows doesnt call it raid, it calls it Dynamic Stripping. You convert any additional drives (cant do it to the boot drive) to dynamic. If you have more than one drive converted to dynamic you can stripe between them (I expect everyone here knows this, but it helps me get it sorted in my mind). This is effect gives you Raid 0. Here is the benchmark:

atto-raid0.jpg

Interesting no? The results IN WindowsXP are as they should be. Maybe a little slower due to my VIA mobo chipset. But why?

Well the two varying factors involved is 1)Dynamic disks 2)Ultra Wide 2 instead of U160.

So to get to the bottom of this I backed up the data on the Seagate Raid 0 setup and coverted them back to basic. They are identical disks so I will only show one benchmark:

This is the Basic NTFS partition:

atto-basic.jpg

This is the Dynamic NTFS partition:

atto-dynamic.jpg

I think the results speaks for itself. If I have just demonstrated the glaringly obvious I apologise, harddrives never been the highlight of my system before now. Before I got scsi I didnt care how fast it went as long is it didnt lose my data. It never affected graphical benchmarks :wink:

What ever the difference is between Basic and Dynamic, this I think is the fix we need.

Share this post


Link to post
Share on other sites

Here are my results:

bench.jpg

I've noticed that in the drive properties page (device manager, windows xp) I don't have the option to enable write cache (it isn't where it has to be, below safe removal options, policies tab)

Is there other way to enable it?

Share this post


Link to post
Share on other sites

I know that this is a stupid question but what program are you using to bench ? I've converted my disks to ynamic in the weekend and didn't bench since then. My that's te cure?

Share this post


Link to post
Share on other sites

Regarding the results for Dynamic disk transfer rates, are these a glitch, or are they a permanent fix? Why would dynamic disks work in such a way? Isn't converting a disk into a dynamic disk Windows XP's way of implementing software RAID? Would this suck CPU power like there' s no tomorrow, nullifying the very reason many of us have SCSI? Do I know what I'm talking about? Can anyone comment on this?

I notice you can create a simple dynamic disc, with no way to switch back; would this affect the entire disk or only a volume? I'd like to have dynamic capabilities in WinXP (if this actually helps) but I also want to test Windows 98's performance characteristics as well...any information? Thanks.

Share this post


Link to post
Share on other sites

Hi,

I was re-setting up my computer today, and I thought I'd investigate this issue.

I put 2 Quantum Atlas 10K II 18.2GB drives on an Adaptec 29160LP, Epox 8KHA+, Athlon XP 1900+, 512MB RAM, 3Com NIC, Radeon gfx card.

I installed Windows 2000 and then Windows XP onto one 10K II, partitioned into two drives, FAT32. The other 10K II was then used for benchmarking - I left it unpartitioned for IOMeter, and as a single empty NTFS volume for all others.

I ran the first few tests both from a clean install (native drivers), and then after updates (SP2, etc. for Win2k) - still native MS drivers for the 29160LP.

I didn't install either the MS patch mentioned elsewhere in this thread or the VIA PCI latency patch, nor any other third party drivers (except the Radeon drivers in 2k).

Here's what I found:

atto-32mb-w2k.jpgatto-32mb-wxp.jpg

:arrow: ATTO benchmarks - the familiar problem in XP, absent in 2k. Mine shows exceptionally bad reads, as per Many_R's results for a 10K II on a 39160. Others show poor writes.

hdtach-w2k.jpghdtach-wxp.jpg

:arrow: HDTach shows similar scores in both 2k and XP (don't know about very high CPU usage anomaly in 2k... and I forgot to do writes)

winbench99-w2k.jpgwinbench99-wxp.jpg

:arrow: WinBench99 also shows no apparent problem. Windows 2000 has a slight edge, but negligble.

iometer.gif

:arrow: I ran the IOMeter test that somebody posted on page 2 of this thread and plotted the transfer rates seen. No problem there either, apparently.

:arrow: I also copied a 768MB file from the first disk (FAT32, not the current system partition) to the empty second disk (NTFS), twice, in Explorer, timing it with a stopwatch. It took 27 seconds (about 28MB/s; little variance) every time, in XP and 2k.

So, there you go, I don't know what to make of it, except that it suggests something up with ATTO, which is the oldest program I think, is it not? :?

Comments most welcome. Sorry about the large and crappy pics...

Share this post


Link to post
Share on other sites

Someone already tried the dynamic disk "fix"? I can't cos I just formatted the drive with WinXP. I'll do a quick install tomorrow to see if it works for me.

Share this post


Link to post
Share on other sites
Regarding the results for Dynamic disk transfer rates, are these a glitch, or are they a permanent fix?  Why would dynamic disks work in such a way?  Isn't converting a disk into a dynamic disk Windows XP's way of implementing software RAID?  Would this suck CPU power like there' s no tomorrow, nullifying the very reason many of us have SCSI?  Do I know what I'm talking about?  Can anyone comment on this?

I notice you can create a simple dynamic disc, with no way to switch back; would this affect the entire disk or only a volume?  I'd like to have dynamic capabilities in WinXP (if this actually helps) but I also want to test Windows 98's performance characteristics as well...any information?  Thanks.

It only creates the software raid when using more than one harddrive. My demonstration is with a single drive, so hopefully not sucking anym ore CPU resources than normal. Dynamic disks ONLY work in Win2K and WinXP.

And no it was NOT a glitch. However XP handles Dynamic disks its alot better than basic.

Obviously this is only a sort of suggested "fix", as you cant have your boot harddrive as a Dynamic disk.

If Microsoft cant get a fix for this and people have to use WinXP maybe the SCSI setup of the future is going to be a "small" 9gb U160 drive and anything additional will be a separate Dynamic drive.

Share this post


Link to post
Share on other sites
Obviously this is only a sort of suggested "fix", as you cant have your boot harddrive as a Dynamic disk.

Are you sure you can't? When I choose "Upgrade to dynamic disk" it tells me "Will Upgrade".

dyn_upgrade.gif

I won't do that now since this is in W2k and I don't have the problem here, but I was going to try with annother SCSI drive tomorrow...

If Microsoft cant get a fix for this and people have to use WinXP maybe the SCSI setup of the future is going to be a "small" 9gb U160 drive and anything additional will be a separate Dynamic drive.

Or rather a small IDE bootdrive? :?

Share this post


Link to post
Share on other sites

atto_xp_3ware7410_GXP75s_RA.gif

This is using 3ware IDE RAID, other spec as gif. I'm not sure if a previous poster was suggesting that IDE RAID cards (as XP lists them as SCSI) are immune, but surely these results suggest that there is something amiss.

Who was it that mentioned feel? I've been using this box for a month now, 2 weeks with the 3ware RAID. I've been gutted at the lack of performance.

Any suggestions (please)?

Share this post


Link to post
Share on other sites

If Microsoft cant get a fix for this and people have to use WinXP maybe the SCSI setup of the future is going to be a "small" 9gb U160 drive and anything additional will be a separate Dynamic drive.

Or rather a small IDE bootdrive? :?

Well I personally find that my SCSI drives still boot faster than my IBM Deskstar.

Share this post


Link to post
Share on other sites

Obviously this is only a sort of suggested "fix", as you cant have your boot harddrive as a Dynamic disk.

Are you sure you can't? When I choose "Upgrade to dynamic disk" it tells me "Will Upgrade".

dyn_upgrade.gif

I won't do that now since this is in W2k and I don't have the problem here, but I was going to try with annother SCSI drive tomorrow...

Oh it will let you alright it just wont be a bootable drive any more. At least thats the way I understand it.

Share this post


Link to post
Share on other sites
Oh it will let you alright it just wont be a bootable drive any more. At least thats the way I understand it.

No, it will boot just fine. The only problem you may encounter is that win95/88 etc won't see the drive. Only win2k/winXP can use it. And you have to convert the whole drive to dynamic, you can't just convert a partition.

Share this post


Link to post
Share on other sites

Actually, WinXP will let you boot from the dynamic cisk, just as long as it is either a simple or a mirrored drive. Any other RAID setups cannot be applied to the boot drive. However, if the entire disk gets converted, I will try to set up W98 on the drive first and do a test there before I try turning it dynamic.

Also, IanGrant, your ATTO benchmarks in W2K look very nice; the Quantum Atlas 10KII is doing what it's supposed to be doing. I was expecting similar (well, doubled) performance as what you have posted, have you tried RAIDing them together? I'd like to see what would come of that...

Share this post


Link to post
Share on other sites

Booting off a dynamic drive? In that case I will test it on my Atlas as soon as I get home from work. Its a new install of WinXP so I am not bothered if it goes tits up. Also I havnt used Win98 since the day Win2k was released.

Share this post


Link to post
Share on other sites
Also, IanGrant, your ATTO benchmarks in W2K look very nice; the Quantum Atlas 10KII is doing what it's supposed to be doing.  I was expecting similar (well, doubled) performance as what you have posted, have you tried RAIDing them together?  I'd like to see what would come of that...

You were expecting double for my results, or you have 2 10KIIs and were expecting double for your results??

Anyway, I used to have them as a RAID 0 volume on an Adaptec AAA-131U2 controller. I did ATTO and HDTach benchmarks for that in XP - they are on page 2 of this thread - not that good, really, but they didn't show the strange v. poor read performance as here. I did bench them long before with WinBench and possibly in 2k, but I can't remember the figures, except that the transfer rate test in WinBench gave a top figure of about 65MB/s (this wasn't in XP, I don't think).

Share this post


Link to post
Share on other sites

IanGrant- great work! Thanks for sharing.

Guess I'll relax (also because the benches I posted for the X15 aren't so bad)

One last question- When I run the bench on Sandra it compares my ALL SCSI system drives to ATA.

WTF is up with that??? Anyway to change?

Sandra sees them as SCSI under its SCSI and system areas....

Share this post


Link to post
Share on other sites

Well, for me at least :)

You may of read in another thread how I`ve managed to lose 40gb across two drives due to dynamic disk. They didnt import for me on a new system..

So with that in mind I was slightly apprehensive in turning main main machiens scsi drive to dynamic, however I can't resist fiddling so I did it.

I thought when I went to xp I lost rather a lot of the snappiness the scsi had given me, but I wrote it off as XP being newer and fatter.

You`ll see below why it felt slow!!

Here is my ATTO before, basic disk ntfs, WinXp Pro on dual P3 700's with an Adaptec 2940UW and Atlas 10kII...

nodynamic.gif

Pretty stinker right? Why some have problems with writes and others reads I don't know, but after converting to dynamic my ATTO looks like this...

dynamic.gif

Yea!!!:) The speeds aren't great compared to newer scsi stuff, but it reads over 3 times faster than before and boy can you notice it!!

I just hope when I hose the install or whatever, I don't have the problem I had with my previous dymanic disk :/

Good luck ppl

Share this post


Link to post
Share on other sites

Well it appears converting the disks to Dynamic have done the trick. Its dropped the scores a tiny bit from Win2K but for someone like me that doesnt want to go back to Win2k its perfect fix. Here is the difference between Win2k and WinXP:

atto-benches-dynamic.jpg

Share this post


Link to post
Share on other sites

Noxis: Feel like running another test with the Dynamic disk under Windows 2000, to compare to the Basic disk Win2k test you already posted?

Share this post


Link to post
Share on other sites

How about running WinBench99, IOMeter, HDTach, and file copy tests on basic and dynamic disks (in 2k and XP)?

I would be grateful if people could try and replicate (or not) my results, as I personally feel something is up with ATTO. It seems Quantum Atlas 10K IIs always show worse reads than writes, while others show worse writes or something else.

Ian.

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now